From Chat-with-PDF to Quiz-Master
summary
Builds a stateful Python app that converts a 130-page PDF into auto-generated multiple-choice and free-text questions using Docling for layout and tables, DeepEval for synthetic QA and distractors, and Marimo for the UI, then grades answers in real time with an LLM-as-judge pipeline. It anchors each QA pair to source coordinates for auditability, handles tables/figures, ran locally with cached models (demo mostly works), notes OCR noise pitfalls and low demo cost (≈$3).
AI-written summary · about
chapters (11)
- 0:00Introduction
- 3:57Unverified RAG answers problem
- 9:13Processing PDFs to text
- 12:25Semantic chunking with Dockling
- 16:19Generating trustworthy questions
- 20:33Deterministic RAG systems
- 22:19OCR noise and sanitization
- 24:30Grading free-text answers
- 28:19Reactive UI with Marimo
- 31:28Live demo: Quiz-Master
- 37:16Key takeaways
related talks
15mTrainSim: A GenAI Restaurant Server Training and Customer Simulation ChatBot
35mVue-doo Magic: AI Development Tricks
21mStructuring the Unstructured
01h 27mBuilding Contextual ChatBot using LLMs, Vector Databases & Python
01h 13mDocument Q&A using LangChain
47mEnd-to-End Chatbot Web App with Your Own Data
32mChatBot Learning Assistant with Large Language Models
01h 00mBuilding an Expert Q&A Bot with Open Source Tools and LLMs